Mathematics, Hybrid computing and HPC

International audienceHPC appears more and more as a key player in the field of numerical simulation and data processing. This trend comes of course with the desire to perform simulations that are closer and closer to real world situations, and with the development of clusters and platforms that provide access to hundreds to thousands CPU/GPU nodes. The application domains encompass many fields, from fluid mechanics to biology and nano-sciences, in academic research as well as for industrial applications. Concerning industrial applications, major groups have often already a good practice of HPC, with dedicated manpower and available in-house platforms. The access of SMEs to HPC is more problematic as they do not have the appropriate resources in hardware and manpower, and it is sometimes hard for them to have a clear idea of the gain they will obtain through HPC. In the first part of the talk, I will talk about a national initiative led by INRIA, GENCI and BPI, to promote the access of SMEs to HPC. This initiative provides support both in terms of market analysis, access to hardware and technical environment. It now involves middle-size HPC platforms that are distributed in French universities. This initiative will therefore give new opportunities to researchers, in particular mathematicians, to be connected to industrial collaborations. HPC is actually not only a question of accessing hardware and adapting existing codes to massively parallel platforms. It also raises questions about mathematical and numerical models that optimize the emerging hardware and analyze the huge amount of data associated with these simulations, and software engineering to distribute algorithms on heterogeneous clusters. Mathematicians therefore can use HPC as a mean to access challenging industrial collaborations in which they can contribute through new methods and algorithms, in both scientific computing and statistics